Large Language Models Enable Effective Deanonymization of Pseudonymous Online Users
By
mellosouls
Reliable enough to start your morning with. Toast it again tomorrow.
Summary
Researchers demonstrate that large language models can effectively perform large-scale deanonymization attacks, re-identifying pseudonymous users across platforms like Hacker News, LinkedIn, and Reddit with high precision. The LLM-based approach extracts identity-relevant features from unstructured text, uses semantic embeddings to find candidate matches, and reasons over top candidates to verify matches. In evaluations, LLM methods achieved up to 68% recall at 90% precision, substantially outperforming classical baselines, showing that practical obscurity for online pseudonymity no longer holds and privacy threat models need reconsideration.
Key quotes
· 4 pulledWith full Internet access, our agent can re-identify Hacker News users and Anthropic Interviewer participants at high precision, given pseudonymous online profiles and conversations alone, matching what would take hours for a dedicated human investigator.
Compared to classical deanonymization work (e.g., on the Netflix prize) that required structured data, our approach works directly on raw user content across arbitrary platforms.
In each setting, LLM-based methods substantially outperform classical baselines, achieving up to 68% recall at 90% precision compared to near 0% for the best non-LLM method.
Our results show that the practical obscurity protecting pseudonymous users online no longer holds and that threat models for online privacy need to be reconsidered.
You might also wanna read
AI System Can Identify Pseudonymous Social Media Accounts, Threatening Online Anonymity
Researchers from ETH Zurich, Anthropic, and ML Alignment Scholars have developed an AI system that can identify pseudonymous social media ac
The Verge·2mo ago
Study finds large language models vulnerable to classic persuasion tactics for harmful requests
This study tested whether three widely used large language models (LLMs) are susceptible to classic persuasion principles (authority, social
pnas.org·4d ago
Study finds LLMs persist in treating false claims as true despite explicit warnings
A study on fine-tuning large language models (LLMs) reveals that even after explicit warnings that certain claims are false, the models cont
arstechnica.com·23h ago
LinkedIn Researchers Propose Unified SLM Framework for Industrial Semantic Search Query Understanding
This paper presents a unified structured query understanding framework for industrial semantic search, developed and deployed at LinkedIn. T
arxiv.org·3d ago